A vocal microphone in front of a vintage analogue synthesizer keyboard with rows of knobs, in a dim studio lit by warm lamps
doodlesynthesizer

HOW A SYNTH LEARNED TO TALK

Speech in,
robot voice out.

VOICE encoder

AI-generated image, not a photograph of a real person or place.

The vocoder treats your voice as a set of moving filters.

It was built in the 1930s to save space on telephone lines, then kept wartime leaders' calls secret, then became the sound of every film robot. The trick behind all three is the same: measure speech in bands, then rebuild it.

Count the bands

THE SHORT ANSWER

Measure the voice,
then rebuild it.

One idea from 1928,
three very different lives.

HOW IT WORKS

A voice as moving filters

A bank of filters splits speech into bands. The loudness in each band steers a matching filter on a buzzy, harmonic-rich tone, so the tone takes on the shape of the words.

Count the bands

WARTIME

Secret calls

From 1943 the SIGSALY system digitised and encrypted vocoder signals for the highest-level Allied calls. The Science Museum says no conversation is recorded as broken.

See the numbers

MUSIC

Robot voices

The classic vocoder loses some pitch inflection and crisp consonants. That made it Hollywood's go-to sound for robots and talking computers, and a studio favourite from the 1970s.

The story
A 1940s green telephone handset resting on a desk beside tall racks of valve electronics in a dim basement room
AI-generated image, not a photograph of a real person or place.

FOUR STAGES OF A VOCODER

From a spoken word
to a singing synth.

A classic channel vocoder has an analysis half and a resynthesis half. In music, the synth you play becomes the voice's new raw sound.

  1. Split the voice

    Speech passes through a parallel bank of bandpass filters, each tuned to a slice of the speech range, from low to high.

  2. Measure each band

    An amplitude detector follows the loudness in each band and turns it into a slowly changing control signal.

  3. Supply a raw sound

    Dudley used a buzzy oscillator rich in partials, standing in for the vibrating vocal folds. Musicians feed in a synth instead.

  4. Rebuild through matching filters

    Each control signal opens a matching filter on the raw sound, so it takes on the moving shape of the original words.

TWO NUMBERS WORTH KNOWING

Very little data,
a very big machine.

One figure on how little a vocoder sends,
one on what it took to keep it secret.

50 a second

Updates per second for intelligible speech

An early modified vocoder updated its control signals only 50 times a second and still produced understandable speech, which is why phone companies were so interested.

DATA · INDIANA UNIVERSITY

55 tons

Weight of the wartime secure phone

SIGSALY filled 2,500 square feet and needed 30,000 watts. Its London terminal sat in the basement of Selfridges because it was too big for Whitehall.

SIGSALY · SCIENCE MUSEUM

Each SIGSALY call was encrypted with random noise stored on a pair of phonograph records, one for each end, each holding 12 minutes of key. The system was in use until 1946.

COUNT THE BANDS

How many bands
does a voice need?

A channel vocoder splits speech into a number of bands, or channels. Pick a band count to see where it was used. The bar shows how much detail the vocoder can carry.

More bands and a wider frequency range give more accurate speech. Musicians chasing the classic effect usually stay at sixteen bands or fewer.

Band check

Choose a number of bands
8 bandsBare minimum

The low end of what practical channel vocoders used. Telephone engineers asked how few bands were just good enough to pass intelligible speech.

10 bandsThe original

Homer Dudley's vocoder used ten bands 300 Hz apart, covering speech from 0 to 2,950 Hz.

16 bandsStudio and military

A 1979 studio vocoder used sixteen bands from 50 Hz to 5,080 Hz. The US HY-2, developed from 1961, was a 16-channel, 2,400 bits-per-second system.

24 bandsUpper end

The top of the usual range of eight to twenty-four. More bands rebuild speech more accurately, but lose some of the robot character.

Sources: Indiana University, Music and Technology text, Speech synthesis and the channel vocoder; NSA, A History of Secure Voice Coding.

Close-up of the front panel of an old analogue audio machine with a row of matching rotary knobs and small sliders, no labels visible
AI-generated image, not a photograph of a real person or place.

FROM BELL LABS TO THE STUDIO

Three lives
of one idea.

Homer Dudley began studying how to imitate the vocal tract electronically at Bell Labs in 1928, and received vocoder patents in 1937 and 1939. His Voder, a keyboard-played speaking machine, was demonstrated at the 1939 World's Fair by operators trained for a year.

After the war, the US military used ever smaller vocoders for encrypted speech until the early 1960s. When affordable vocoders reached studios in the 1970s, musicians used them to make voices sing through synths. The Science Museum holds a UK kit-built vocoder used by an electro-acoustic musician.

In conversation

Questions people ask

What does vocoder stand for?

It is a contraction of voice encoder. Homer Dudley developed it at Bell Labs, receiving patents in 1937 and 1939.

How does a vocoder make a robot voice?

It measures the loudness of speech in a set of frequency bands and uses those measurements to shape another sound, such as a synth. The classic design loses some pitch inflection and crisp consonants, which gives the robotic character.

Was the vocoder really used in the Second World War?

Yes. The SIGSALY system, introduced in 1943, digitised and encrypted vocoder signals for top-level Allied calls, including between Churchill and Roosevelt. It weighed 55 tons.

Is vocoder technology still used?

Yes. Indiana University's music technology text notes that vocoder technology is still employed in digital forms of modern telephony, and vocoders remain available as audio plug-ins.

What was the Voder?

A speaking machine Dudley built using the resynthesis half of his vocoder, played by hand with ten keys, wrist bars and a foot pedal. It was shown at the 1939 World's Fair.

Sources

Each figure on this page comes from the sources below.

  1. Indiana University, Music and Technology text: The channel vocoder, early electronic speech synthesis
  2. Indiana University, Music and Technology text: How the channel vocoder works
  3. Indiana University, Music and Technology text: A brief history of the vocoder
  4. Science Museum blog: Scrambled phones
  5. NSA: A History of Secure Voice Coding (PDF)
  6. Science Museum Group Collection: Electronics Today International vocoder by Powertran

ABOUT THIS NAME

Interested in leasing doodlesynthesizer.com?

doodlesynthesizer.com is part of the DoodleSale portfolio. It isn't in an auction right now, but you can ask about leasing it.

Ask about this name